All tags

Topic: "autoregressive-models"

    DeepSeek-OCR finds vision models can decode 10x more efficiently with ~97% accuracy of text-only, 33/200k pages/day/A100
    Gemini 2.5 Pro + 4o Native Image Gen